Search Techniques for Automated Proposal of Data Mining Schemes

0470347 - ÚI 2017 RIV CH eng C - Konferenční příspěvek (zahraniční konf.)
Neruda, Roman
Search Techniques for Automated Proposal of Data Mining Schemes.
Applied Computer Sciences in Engineering. Cham: Springer, 2016 - (Figueroa-García, J.; López-Santana, E.; Ferro-Escobar, R.), s. 84-90. Communications in Computer and Information Science, 657. ISBN 978-3-319-50879-5. ISSN 1865-0929.
[WEA 2016. Workshop on Engineering Applications /3./. Bogota (CO), 21.09.2016-23.09.2016]
Grant CEP: GA ČR GA15-19877S
Institucionální podpora: RVO:67985807
Klíčová slova: computational intelligence * machine learning * meta-learning
Obor OECD: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Data mining schemes, or workflows, are collections of interconnected machine learning models, including preprocessing procedures, and ensembles methods combinations. The proposal of data mining schemes for a task at hand has always been a task for experienced data scientists. We will study generating and testing workflows by automated procedures. Two representations of data mining schemes are used in this paper - a linear one, and a one based on direct acyclic graphs. Efficient procedures for generating schemes are presented and evaluated by testing the generated schemes on real data.
Trvalý link: http://hdl.handle.net/11104/0268011